Past studies have danced around the notion that nonsteroidal anti-inflammatory drug use might carry risks when utilized in certain infections. This most recent study is well-designed and performed, but what do we do with the results?
Source: LeBourgeois J, Ferroni A, Leruez-Ville M, et al. Nonsteroidal Anti-Inflammatory Drug without Antibiotics for Acute Viral Infection Increases the Empyema Risk in Children:A Matched Case-Control Study. J Pediatr 2016;175:47-53; doi:10.1016/j.jpeds.2016.05.025. See AAP Grand Rounds commentary by Dr. Daniel Lesser (subscription required).
This case control study, performed in France with a number of features to limit bias in the results, found that children with acute viral infection who received at least 1 day of nonsteroidal anti-inflammatory drug (NSAID) treatment had a higher risk of empyema compared to those who did not; specifically, children who subsequently developed empyema (cases) had significantly increased odds of being exposed to at least 1 day of NSAIDs than did those who didn't develop empyema (controls), with an adjusted odds ratio of 2.79 (95% confidence interval 1.40 - 5.58). Antibiotic use in children exposed to NSAIDs was associated with a lower risk of empyema, compared to those not exposed to NSAIDs.
As I said above, this study is very well-designed, and I recommend this article to anyone planning a case control study. However, I want to focus on 1 aspect of this design that was particularly clever in its attempt to minimize the possibility of protopathic bias, a particularly tough problem in case control studies. This occurs because, prior to identification of individuals as "cases," in this case children who ultimately developed empyema, something about their initial presentation made them more likely to receive the intervention in question, NSAID therapy. In other words, did they already appear somehow "sicker," such that they were managed differently?
If that was the case, it was really their initial presentation, rather than the NSAID administration, that led to empyema. Remember, case control studies can only demonstrate an association of factors, which isn't the same as showing cause and effect. These researchers minimized protopathic bias by pairing cases and controls from the same practice/physician setting, so that management strategies might have been more homogeneous. Also, they excluded children from analysis if they demonstrated 2 conditions that could be prone to introduce protopathic bias: 1) children who didn't have at least 24 hours of afebrility between recovery from their initial viral illness and the diagnosis of empyema; and 2) the time between viral illness onset and diagnosis of empyema was less than 72 hours. Furthermore, children included in the analysis must have had NSAID use within 72 hours after the onset of the initial viral syndrome, avoiding inclusion of children with prolonged illnesses.
But back to our main question, what do we do with these findings? In particular, what the heck does an odds ratio of 2.79 really mean? For most situations, the odds ratio number doesn't translate easily into a clinically useful estimate of how many patients might benefit from a change in strategy such as not using NSAIDs for viral respiratory infections. I already stated that case control studies show an association, which is not the same as stating that NSAID use causes empyema in this example. However, if we do assume that this is reflecting a cause and effect mechanism, can we estimate how commonly this happens?
Mathematical manipulations are available, the most accessible being those from the Centre for Evidence Based Medicine at University of Oxford. Scroll to the bottom of their page on number needed to treat, and you'll see a conversion table for this purpose. The problem with going through these mental gymnastics is that, in our example, we're dealing with an extremely rare event, i.e. the number of children with a runny nose who end up with pulmonary empyema. I couldn't even find a reliable estimate of this incidence, but it certainly is far less than 0.1%, probably by orders of magnitude. In other words, spending a lot of effort abolishing NSAID use for colds isn't likely to alter the rates of pulmonary empyema very much. So, I think we should all sit tight for now until someone does the randomized prospective trial to give us more proof of a causal nature for empyema with NSAID use in viral respiratory infections.